Remote sensing techniques have been applied extensively in geospatial investigations, but their use in measuring soil physical attributes has been far less explored. Soil particle size distributions (PSD) are indispensable in modeling pedological and hydrological processes as well as biodiversity. However, estimation of PSD via gravimetric measurement methods, the standard currently in use, is relatively laborious and time-consuming. This research is a pioneering attempt to estimate soil PSD from computerized tomographic (CT) scans. CT scanners non-invasively penetrate three-dimensional (3D) objects to produce a series of two-dimensional (2D) gray images, where grayscale values express density of internal matter in Hounsfield Units (HU). In this study, a model is developed that associates particle size with HU-derived pixels by first classifying the image with an unsupervised technique and then by hierarchically clustering the classes according to soil horizons. The soil PSD is computed as the relative class frequency of classified pixels. For the type of soil used in this study, Weibull distribution was found to fit all layers at a fine 10 mm scale, but a broader horizon-level analysis found lognormal distribution to perform best (in the absence of Weibull). The PSD estimated from CT scans was insignificantly different from the sieve-analysis measured PSD in each horizon. This novel approach to soil diagnostics could transform future soil particle analyses.